🦙 KAGGLE_OPENVACCINE_MRNA 🦙

✍ Authors:

BEL Alexis - BELAKTIB Anas - OUSSAREN Mohamed - ROUAUD Lucas

Master 2 Bio-informatics at Univerité de Paris.

🔎 Interesting path

• 📑 Report: doc/report/report.pdf
• 📢 Oral presentation: doc/presentation
• 🖥 Main: src/main.py

🤔 Context

This project is actually trying to answer to this Kaggle project: OpenVaccine: COVID-19 mRNA Vaccine Degradation Prediction.

The main problematic is the synthesis of a stable mRNA vaccine. Because of the molecular nature of it, the vaccine degraded itself easily and quickly. To counter that, it is necessary to synthesize a mRNA stable.

To do so, the product mRNA have to be tested. And this is where this project take place: creating a neural network to predict the stability of a given sequence.

🧐 Methods Implemented

Because this project is done to validate a course, one mandatory criterion is create two neural network approches. Here, we've done:

• Three different embedding:
  • One from keras.
  • One recode by ourselves.
  • One using RNABERT transformer.
• Two neural networks:
  • One Convutional Neural Network (CNN).
  • One Google Inception.

🚀 Launching this program

🐍 Conda environment

To use this program, you will need to create a conda environment like so:

mamba env create --file kaggle_reseau.yml
conda env create --file kaggle_reseau.yaml
conda activate reseau

⚙️ General method

To launch this program, simply use the next commands (after the activation of the conda environment):

python3 src/main.py --help

🔎 Parameters description

Next, the parameters are described:

Parameters	Parameters name	Usage
*-i, --input	Input X + Y data/neural network trained	Add an .npy data file or a .h5 neural network file.
*-o, --output	Output data Y/neural network finish to be trained.	Add an .npy data file or a .h5 neural network file.
-pred, --predict_data	Output predicted Y data.	Add an .npy data file.
--cnn	Convolutional Neuronal Network.	Add like a True.
--ginc, --google_inception	Google inception's neural network.	Add like a True.
--ke, --keras_embedding	Using classical keras embedding method.	Add like a True.
--owe, --own_embedding	Using our compute pre-embedding.	Add like a True.
--re, --rnabert_embedding	Using embedding compute by RNABERT transformer.	Add like a True.

🧠 One program, two usage

If you don't use the parameters -pred, --predict_data:

You actually said to the program that you want to train neural network. To do that, give to -i, --input a dataset to learn and to -o, --output the neural network to reuse. Do not forget to indicate a type of neural network (-ginc, --google_inception or -ke, --keras_embedding) to use and a type of an input embedding (-ke, --keras_embedding, -hme, --homemade_embedding or -re, --rnabert_embedding).

If you use the parameters -pred, --predict_data:

You actually said to the program that you already have a train neural network. So you want to predict Y data base on X data. To do that, give to -i, --input a trained neural network, to -o, --output how to write the Y predict data and to -pred, --predict_data the X data. Do not forget to indicate the good input embedding (-ke, --keras_embedding, -hme, --homemade_embedding or -re, --rnabert_embedding).